JEQ Journal of Natural Resources and Life Sciences Education
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via ISI Web of Science (14)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Lee, J.-Y.
Right arrow Articles by Lee, M.-H.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Lee, J.-Y.
Right arrow Articles by Lee, M.-H.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Lee, J.-Y.
Right arrow Articles by Lee, M.-H.
Related Collections
Right arrow Ground Water Quality
Right arrow Water Quality
Right arrow Water Pollution
Journal of Environmental Quality 30:1548-1563 (2001)
© 2001 American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America

TECHNICAL REPORT
Ground Water Quality

Statistical Evaluation of Geochemical Parameter Distribution in a Ground Water System Contaminated with Petroleum Hydrocarbons

Jin-Yong Leea, Jeong-Yong Cheonb, Kang-Kun Lee*,b, Seok-Young Leec and Min-Hyo Leec

a GeoGreen 21 Co., Ltd., Research Park Innovation Center 412, Seoul National University, Seoul 151-818, Korea
b School of Earth and Environmental Sciences, Seoul National Univ., Seoul 151-742, Korea
c Soil Environment Division, National Institute of Environmental Research, Incheon City, Korea

* Corresponding author (kklee{at}snu.ac.kr)

Received for publication August 14, 2000.

   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 CONCLUSIONS AND DISCUSSION
 REFERENCES
 
A shallow-depth ground water area was investigated to identify the dominant processes governing the distribution of hydrocarbon contaminants and hydrogeochemical parameters. The ground water in the study site has been highly contaminated with petroleum hydrocarbons. A preliminary pump-and-treatment remediation technology was applied for 4 yr at the site. Multivariate analyses were applied to hydrogeochemical data obtained before and after the rainy season. The pump-and-treatment application, indigenous biodegradation, and mixing by precipitation recharge are the major factors or events involved in the distribution of geochemical parameters of the ground water in the study area. Site-specific artificial pavement also played an important role in the evolution of the ground water chemistry. A conventional graphical analysis method (Piper plot) of major ions did not effectively reveal these effects. In this study, we demonstrate the usefulness of multivariate analysis (factor and cluster analyses) using biodegradation indicator parameters, as well as major cations and anions, for the study of the ground water system in the hydrocarbon-contaminated site.

Abbreviations: BTEX, benzene, toluene, ethylbenzene, and xylene • DO, dissolved oxygen • EC, electrical conductivity • ORP, oxidation–reduction potential • TEX, toluene, ethylbenzene, and xylene


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 CONCLUSIONS AND DISCUSSION
 REFERENCES
 
CONVENTIONAL STUDIES of ground water have placed a heavy emphasis on the variations in the chemical characteristics of ground water in time and space (Kennedy et al., 1999). Therefore, many researchers have performed multiple ground water samplings and subsequent chemical analyses. The main tools for the interpretation of chemical analysis results are graphical methods (usually based on Stiff and Piper diagrams) combined with basic statistics (e.g., average, frequency, correlation) (Montgomery et al., 1987; Hem, 1992; Frapporti et al., 1993). Based on the chemical evolution pattern of natural ground water by water–rock interactions, the major chemical components have been used to characterize the water types. However, interpretations based on the above methods using major chemical species may be misleading or ineffective for a petroleum-contaminated ground water system, because there are other important parameters representing dynamic hydrogeochemical conditions.

For a better understanding of the hydrogeochemistry of the ground water system, multivariate analyses can be performed using not only concentrations of chemical species (major cations and anions, redox-related species), but also other physicochemical data such as temperature, redox potential, dissolved oxygen, electrical conductivity, pH, and alkalinity. As indicated in Suk and Lee (1999), spatial or temporal measurements of chemical or physical properties usually do not directly reveal the underlying governing processes in the ground water system of interest. These authors used factor and cluster analyses to interpret the spatial distribution and temporal evolution patterns of many chemical and physical parameters, without significant loss of measurement and chemical analysis data, and without obscuring the geochemical meaning of the data. Factor and cluster analyses have been employed to reveal the most important governing processes and the hydrogeochemical similarities between the observation points through data reduction and classification (Suk and Lee, 1999). Several researchers (e.g., Usunoff and Guzman-Guzman, 1989; Ritzi et al., 1993; Ochsenkühn et al., 1997) have applied factor and/or cluster analyses to ground water chemical data in order to understand ground water systems.

The present study monitored the chemical characteristics of ground waters in a petroleum-contaminated site. Before the study, a pump-and-treatment system had been applied, causing artificial disturbances in the physical and chemical ground water systems. Unfortunately, the pump-and-treatment system was operated without detailed characterization of the ground water flow system and, more importantly, without contamination source control at the site. As a result, the extent of ground water contamination seemed to have changed, and the shape of the plume had become irregular, with multiple centers (Lee, 2000). The pump-and-treatment processes greatly disturbed the hydrochemical conditions of the ground water in this area.

The objectives of this study were to (i) investigate the distribution of petroleum contaminants and hydrogeochemical parameters in ground water through multivariate analysis and (ii) interpret the mechanisms or factors that caused the present hydrogeochemical conditions. The major concern in the interpretation of the mechanisms is the effect of more than 4 yr of pump-and-treatment practices on the hydrogeochemical parameters. In this study, chemical and physical parameters including benzene, toluene, ethylbenzene, and xylene (BTEX) concentrations, concentrations of 17 chemical species, and field measurement data (water level, temperature, dissolved oxygen, redox potential, pH, and electrical conductivity) were collected during two rounds of sampling from monitoring wells at the petroleum-contaminated site.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 CONCLUSIONS AND DISCUSSION
 REFERENCES
 
Site Characteristics and Hydrogeology
The study site is located about 45 km southeast of Seoul, Korea (Fig. 1). Subsurface contamination by petroleum hydrocarbons was first detected at this site when water from a shallow ground water well, installed in the center of the site, became undrinkable in March 1983 (Lee et al., 2001a,b). After that time, the well was abandoned. In September 1990, the grasses in the area within about 10 to 20 m from the petroleum storage tanks died. This incident caught the attention of manufacturers in the contaminated area. Two companies, a paint manufacturer and a textile company, are located in the area. The aboveground and underground storage tanks of the paint manufacturer are located near the textile property (see Fig. 1).



View larger version (38K):
[in this window]
[in a new window]
  		Fig. 1. Location of the study site, and the layout of the monitoring wells and the piezometers. The location of the petroleum hydrocarbon storage tanks is also shown.

 
As a preliminary corrective measure to the ground water contamination problem, a pump-and-treatment remedial system was installed. The system includes six ground water pumping wells that were in operation from June 1994 to October 1998. Ground water was pumped from W-series wells (W1–W6; see well locations in Fig. 1) to remove free products and dissolved hydrocarbons from the subsurface. In more than 4 yr of operation, about 2350 kg of petroleum hydrocarbons were removed by pumping 24900 m3 of ground water.

Chemical analyses revealed that the main components of the dissolved plume were toluene, ethylbenzene, and xylene (TEX), which are organic solvents used for industrial paint production. The company that manufactures the industrial paint has been operating on this site since 1973. The manufacturing company is still in operation at present. In the past, the company has used 500000 to 800000 kg of toluene, 1100000 to 1400000 kg of xylene, and 250000 to 350000 kg of other solvents each year for paint production. These liquids have been stored in above and underground storage tanks. Among them, three aboveground storage tanks have the largest volume of all storage tanks. The three tanks, with a storage capacity of 140000 L, have been used to store one of diesel, xylene, toluene, Bunker A (a hard residual fuel oil), or Bunker C (a heavy residual fuel oil). These storage tanks are most likely the primary source of the extensive subsurface contamination in this area.

Geological information on the study site was obtained from the logging data of 87 piezometers and monitoring wells. Stratigraphic units underlying the site include Precambrian gneiss, overlying alluvial deposits and local reclamation soil (Fig. 2). The upper part of the Precambrian gneiss is slightly weathered and lies 5 to 6 m below the ground surface. This is practically the bottom of the shallow aquifer in this area. The alluvial deposits are mainly composed of two layers: fine sand and gravelly coarse sand. The sediments of the two layers are relatively well extended across the whole study area (see Fig. 2). The fine sand and gravelly coarse sand layers range from 0.5 to 2 m, and from 2 to 6 m, respectively, with varying thickness. The two layers are at their thickest in the northwestern part of the study area and thinnest in the southeast. The surface material is either an artificial reclamation soil (clayey-organic soil) or pavement. Portions of the area in the study site were paved, especially in the downgradient area. The main flow of ground water occurs in the gravelly coarse sand layer.



View larger version (26K):
[in this window]
[in a new window]
  		Fig. 2. Hydrogeologic section showing vertical well locations and screen intervals. Locations of the wells are not aligned in the straight line.

 
The water table, at a level of 1.5 to 3.5 m below the ground surface, fluctuates with a vertical range of approximately 2 m over the year, due to precipitation events and ground water flow induced by regional recharge in the upgradient area. The main direction of ground water flow is toward the northwest (Fig. 3). The hydraulic gradient is about 0.006, with seasonal variations. This value increases to 0.014 in the summer. Annual rainfall in this area is about 1300 mm, with more than 60% of the total amount falling during the summer (June, July, and August).



View larger version (54K):
[in this window]
[in a new window]
  		Fig. 3. Ground water level contours for (a) May 1999 and (b) September 1999. Numbers indicate ground water levels above mean sea level in meters.

 
Hydraulic conductivities of the hydrogeologic units were estimated using 70 slug tests, two tracer tests, two short-term pumping tests, and 75 sieve analyses. The gravelly coarse sand layer shows the highest values of hydraulic conductivity, ranging from 5.0 x 10-2 to 1.85 x 10-1 cm/s. The fine sand layer has lower values ranging from 1.5 x 10-3 to 7.6 x 10-3 cm/s. The reclaimed soil layer has the lowest values, being below 10-4 cm/s. Based on the average water table gradient and effective porosity of 0.33, estimated from the two tracer tests, the ground water velocity in the main coarse sand aquifer was estimated to range from 286 to 1058 m/yr.

Chemical Analysis
To evaluate the chemical composition of the ground water in the study area, two rounds of water sampling were performed (May and September of 1999). At least three well volumes of water were purged before sampling, using a low-rate submersible pump attached to a polyethylene (PE) hose. The hose was connected to a closed flow-through cell. Prior to water sampling for laboratory analysis, temperature, pH, electrical conductivity (EC), oxidation–reduction potential (ORP), and dissolved oxygen (DO) were measured with standard probes in the cell. Samples were collected only when the values stabilized. Ferrous iron (Fe2+) was determined in the field using a spectrophotometer (DR2010; Hach, Loveland, CO). Water samples for BTEX analysis were collected directly from the unbroken water stream in 40-mL glass amber vials with Teflon-lined septa and no head space. Samples for multielement analysis were syringe-filtered at 0.45 µm and preserved using ultrapure HNO3 in 125-mL HDPE bottles. Samples for laboratory analysis of anions were collected in 60-mL HDPE bottles through a 0.45-µm syringe filter. Water samples of 125 mL were also collected for laboratory analysis of alkalinity. All samples were stored at 4°C until analysis. More than 120 ground water samples for laboratory analysis of organic and inorganic constituents were collected from the monitoring wells within 2 d for each sampling round. During the sampling periods, there was no rainfall. At least 10% of the samples were spiked and duplicated (Fetter, 1991).

Ionic constituents (NO-3, NO-2, NH+4, PO3-4, SO2-4, F-, Cl-, and Br-) were analyzed by ion chromatography (IC) (DX-120; Dionex, Sunnyvale, CA); other constituents (Al, Fe, K, Ca, Mn, Si, Mg, and Na) were analyzed by inductively coupled plasma–mass spectrometry (ICP–MS) (Ultramass 700; Varian, Palo Alto, CA), atomic absorption (AA) (5100PC; Perkin-Elmer, Norwalk, CT), and ICP (ICP-IIIS; Shimadzu, Kyoto, Japan) following USEPA standard methods. Organic compounds including BTEX were analyzed by gas chromatography–mass selective detection (GC–MSD) with a purge-and-trap system (HP5890, Tekmar3000; Hewlett–Packard, Palo Alto, CA). Alkalinity was determined by potentiometric titration using Gran plots for graphical determination of the end point (Stumm and Morgan, 1996; Drever, 1997).

Multivariate Analysis
Factor analysis attempts to simplify the complex and diverse relationships that exist among a set of observed variables by revealing common and unobservable factors that link together the seemingly unrelated variables (Usunoff and Guzman-Guzman, 1989; Evans et al., 1996). In hydrochemical studies, the results of ground water chemical analysis and field measurement data are the observable variables, and the underlying physicochemical and/or biological processes in the ground water system are the so-called unobservable common factors.

Cluster analysis groups the whole ground water system into a finite number of clusters. Each cluster represents a specific and similar hydrogeochemical state of ground water. Usually, cluster analysis is applied to the raw data (the observable variables, see Frapporti et al., 1993; Ochsenkühn et al., 1997). In this study, the factor scores were used instead of the raw data, in order to prevent the cluster analysis from involving unnecessary or trivial factors and mutually dependent variables (Suk and Lee, 1999). For the multivariate analyses, the SAS package (SAS Institute, 1997) was used.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
CONCLUSIONS AND DISCUSSION
 REFERENCES
 
Chemical Analysis
A summary of the results of the chemical analyses and the field measurement data for each of the two sampling rounds (May and September of 1999) is presented in Table 1. Commonly identified petroleum hydrocarbons are toluene, ethylbenzene, and xylene (TEX), which are major solvents for industrial paint. Benzene concentrations in the ground water are below the detection limit (Lee et al., 2000). The high concentrations of total TEX were detected in the vicinity of the storage tanks and in the center of the study site (Fig. 4). The highest concentrations were found at wells LM28 (100–137 mg/L) and LM8 (35–78 mg/L) for May and September sampling. The locations of the TEX hot spots did not change over the sampling interval. Ground water from a well (HM1) upgradient from the contamination source was taken to measure reference characteristics of the ground water in this study. However, this reference ground water was also slightly contaminated and so it does not provide true background chemistry. The ground water at this well contains low levels of dissolved oxygen (~1.5 mg/L), moderate levels of nitrate (~3.0 mg/L as N), and high levels of sulfate (150–180 mg/L). Dissolved iron and manganese at this well are very low (<0.2 mg/L for Fe and <0.02 mg/L for Mn). The ground water is near neutral (pH ~6.8) with a low buffering capacity (alkalinity ~1.8 meq/L). The oxidation–reduction potential (ORP) is positive (~100 mV).


View this table:
[in this window]
[in a new window]
  		Table 1. Statistics of physicochemical parameters of ground waters sampled in May and September 1999.

 


View larger version (57K):
[in this window]
[in a new window]
  		Fig. 4. Distribution of total toluene, ethylbenzene, and xylene (TEX) concentrations in ground water for (a) May 1999 and (b) September 1999. Units are mg/L.

 
Average concentrations of total TEX in the study area were approximately 19 and 12 mg/L for the May and September sampling, respectively. Generally, dissolved oxygen was depleted (<1.0 mg/L) as the primary electron acceptor in hydrocarbon degradation (Chiang et al., 1989; Cozzarelli et al., 1994; Chapelle et al., 1995). Nitrate concentrations in the wells downgradient from HM1 were very low (< 0.1 mg/L). There was essentially no change in oxygen and nitrate during transport from the source area to about 600 m downgradient since both electron acceptors were already depleted, which indicates that aerobic respiration and denitrification had occurred. The large decline in sulfate (by at least 100 mg/L) and increase in dissolved iron (by at least 40 mg/L) in wells downgradient relative to the reference concentrations (HM1), indicate that both sulfate and iron reduction are occurring. The decline in redox potential relative to HM1 reflects the change from oxidizing conditions upgradient of the source area to reducing conditions in the downgradient area (Wiedemeier et al., 1996). Elevated concentrations of manganese also indicate anaerobic manganese reduction.

Paired t-tests suggest that the hydrogeochemical conditions of ground water samples collected in May and September are globally different (the two means of 15 out of 23 parameters are significantly different at the P = 0.05 level). Changes in the parameters in each well reveal that some physicochemical and/or biological processes progressed during the sampling interval (Fig. 5). A decline in the concentrations of total TEX by about 35% in all the monitoring wells (except for LM20, ~3% reduction) had occurred by September. The nearly uniform reduction of total TEX is due to the mixing and dilution of the ground water by recharge due to a large amount of rain falling over a short time (760 mm during June, July, and August), and to a certain extent, further biodegradation (Lee et al., 2001b). This was quite consistent with our expectations. This explanation is well supported by a large decline in conservative species of chloride (Cl-) and bromide (Br-) in all the wells (Cheon, 2000). Most likely, Fe2+ was reprecipitated due to the large influx of O2–rich recharge in the upgradient area, resulting in the decline of Fe. Nitrate in most of the wells decreased, the largest decline detected at S23. However, concentrations of nitrate at HM1 and M2 increased by 3 to 8 mg/L (as N), which can be attributed to water recharge by rain. Almost the same is true for sulfate. Concentrations of sulfate in most of the downgradient wells decreased. Dissolved oxygen decreased in all the wells monitored by up to 1 mg/L. The ORPs in this area varied in a distinctive manner. The values of ORPs increased by up to 200 mV in upgradient wells, while the values decreased by up to 400 mV in downgradient wells. This implies that ORPs at upgradient wells are directly affected by water recharge.



View larger version (31K):
[in this window]
[in a new window]
  		Fig. 5. Changes in the geochemical parameters between May and September 1999 at monitoring wells along the main direction of ground water flow.

 
Based on the results of the above analysis, it can be inferred that variations and distributions of the geochemical parameters in May were influenced by hydrocarbon degradation processes, while in September they were mainly affected by recharge from rainfall occurring during the wet season, especially in the upgradient area. However, while a biological reduction of iron, sulfate, and nitrate in the downgradient area cannot be ignored, it cannot be readily separated from dilution by ground water recharge (Lee et al., 2001b). Furthermore, large differences in geochemistry occurring over a short time period (90 d), or over a short distance, are not uncommon. The large influx of new water involves contaminant dilution, iron precipitation, further biodegradation, accelerated transport of chemical species, and supply of depleted electron acceptors. They can occur in a shallow aquifer where rainfall recharge significantly affects water chemistry (e.g., Vroblesky and Chapelle, 1994).

Piper Plots
The Piper plot has been a traditional method of classification in the study of hydrochemistry (Ophori and Tóth, 1989; Hem, 1992). It can easily be shown that the analysis of any mixture of waters will plot in a straight line (Hem, 1992). However, the method has limited usage because of the selection of available parameters (Ca, Mg, Na, K, HCO3, Cl, and SO4) and an arbitrary choice of classification limits (Frapporti et al., 1993). The geochemical data in this study, plotted as Piper diagrams, are presented in Fig. 6. Most of the ground water is classified as calcium–bicarbonate (Ca–HCO3) type, a dominant water type that prevails in most petroleum-contaminated waters. Only at the periphery of the contamination plume (M2, M4, S23), or in an upgradient well (HM1), do different types appear (Ca–SO4–HCO3 or Ca–Cl–HCO3 types). However, the classification based on the limits within the Piper diagram (encircled in Fig. 6), generated a markedly different grouping or zonation from the results of the cluster analyses (discussed later). The different grouping is mainly due to the limited use of geochemical data (only major constituents) in the Piper plot. A comparison of the Piper plot for September with that for May reveals that most of the ground waters in this area underwent changes in their major chemical compositions. A distinctive feature shown in the Piper plot for September is that most ground waters became homogeneous. This homogenization process was mainly due to mixing with newly recharged water from the large amount of rainfall during this period. From the results of the Piper plots, it can be inferred that this type of classification is not suitable for distinguishing geochemically homogeneous groups (similar redox conditions) representing similar ongoing biodegradation processes (Frapporti et al., 1993).



View larger version (18K):
[in this window]
[in a new window]
  		Fig. 6. Piper plots showing major chemical compositions of the ground waters for (a) May and (b) September 1999.

 
Factor Analysis
An exploratory inventory of the ground water geochemistry data was performed prior to an initial factor analysis. The Shapiro–Wilk normality tests (Shapiro and Wilk, 1965) and box plots were included in this investigation to identify abnormal values or outliers. The Shapiro–Wilk tests showed that the parameters of interest approached a normal distribution at P = 0.05 without the log transformation, which has been commonly applied to hydrochemistry data (e.g., Frapporti et al., 1993; Ochsenkühn et al., 1997; Reimann and Filzmoser, 2000; Carlon et al., 2001). Based on the exploratory data analysis, the data for each parameter was standardized and scaled to equalize the influence of parameters with small and large variations using the z transformation (Auf der Heyde, 1990; Ravichandran et al., 1996; Ochsenkühn et al., 1997). The calculated z scores were used as input for the factor analysis. Twenty-one variables of a total of 23 observed parameters were adopted for initial factor analysis. The water level was excluded because of a very small value of the coefficient of variation. Ferrous iron (Fe2+) was also excluded because, for both sampling times, the values were determined in the field 1 wk after water sampling. However, it appears that the ferrous iron (Fe2+) data are well represented by total dissolved iron (Fe) (correlation 0.94 at P < 0.0001 in this study) in the nearly neutral pH conditions (Hem, 1992; Kennedy et al., 1999).

Table 2 shows the correlation matrix of z scores for the 21 variables from the two sampling periods. The correlation patterns for both sampling times do not differ markedly. In particular, the concentrations of total TEX do not show any significant correlation with the other parameters for either sampling period. In fact, a meaningful correlation can be expected between the initial TEX concentration (not the final TEX) and the final biodegradation indicator parameters (Chiang et al., 1989). However, there are no data on the initial TEX concentrations; nor can the concentrations be readily estimated. Significant correlations over 0.80 were found between Ca and Mg, Na and SO4, SO4 and DO, and alkalinity and electrical conductivity (EC). The negative correlation of -0.67 between ORP and dissolved iron (Fe) suggests that the increase in Fe concentration is accompanied by the decrease in ORP.


View this table:
[in this window]
[in a new window]
  		Table 2. Correlation among the ground water chemistry and field measurement data (May and September 1999).

 
Based on the computed correlations between the 21 variables, and in the context of geochemical importance, only 15 variables (Ca, Fe, Mg, Mn, K, Si, Na, NO3, SO4, alkalinity, temperature, EC, pH, DO, and ORP) were selected for the final factor analysis. The selection of a limited set of variables should avoid the problems associated with too many factors, which might complicate the hydrogeochemical interpretation of the ground water. The factor analysis result using 21 variables indicated that 9 factors were needed to explain about 90% of the total variance, which was explained by only 5 factors in the case of using 15 variables. Furthermore, the final communality was not appropriately assigned to each variable.

Table 3 shows the eigenvalues of the extracted factors, the eigenvalue difference among factors, and the proportion of total sample variance explained by the factors. The first five factors were selected to represent the hydrogeochemical and/or biological processes of the ground water, without losing significant information. The selection of the five most important factors is based on the eigenvalue criterion (>1) and the variance explained by the extracted factors. The first five factors explain about 90% of the total sample variances for both sampling periods. For a simpler and easier interpretation, factor rotation was performed using varimax rotation after factor extraction.


View this table:
[in this window]
[in a new window]
  		Table 3. Eigenvalues of factors extracted through principal component analysis, differences between factors, and proportion of variance explained by the factors (May and September 1999).

 
Table 4 shows the rotated factor pattern of five extracted factors for two sampling events. Factor 1 for May explains the largest proportion (39%) of the total variance. There are high positive loadings for potassium (K), sodium (Na), sulfate (SO4), and dissolved oxygen (DO). This factor is associated with ground waters with high concentrations of K, Na, SO4, and DO. In other words, Factor 1 is globally indicative of the hydrochemistry of areas with less hydrocarbon contamination, or of areas recently recharged by rain (Frapporti et al., 1993). Water with high concentrations of sulfate (SO4) and dissolved oxygen (DO) may originate from an upper part of the source area, as shown by the high loading of Factor 1 for well HM1. Factor 2 accounts for the major part of the variances of Ca, Mg, alkalinity, Si, and EC. Alkalinity and hardness (as Ca + Mg) would increase in an area affected by the petroleum hydrocarbons (Cozzarelli et al., 1995). The excess of alkalinity relative to calcium (Ca) is likely to be derived from biodegradation of petroleum hydrocarbons (Borden et al., 1995; Basberg et al., 1998; see Fig. 7). It seems probable that most of the dissolved silicon (Si) observed in the ground water originally results from the dissolution of silicate minerals in processes of weathering (Hem, 1992). Multiple geochemical processes involving the increases of alkalinity, hardness, and electrical conductivity can largely characterize Factor 2.


View this table:
[in this window]
[in a new window]
  		Table 4. Rotated factor pattern of five extracted factors after varimax rotation.

 


View larger version (20K):
[in this window]
[in a new window]
  		Fig. 7. Plot of alkalinity versus calcium with 1:1 alkalinity–Ca line (dotted).

 
Factor 3 has high positive loadings for Fe, Mn, alkalinity, and EC, and a moderate negative loading for ORP. The first four parameters (Fe, Mn, alkalinity, and EC) present the highest concentrations or values in the most reduced zone (HM2, HM3, and HM4). Factor 3 is mostly associated with the increase in iron and manganese through reduction. The highest loadings of Factor 3 are located at HM2, HM3, and HM4. Factor 4 is associated with ground waters with high concentrations of NO3 and high values of temperature. High water temperatures are largely associated with high water tables, which relate to the water recharged from rain. Factor 4 also represents the opposing geochemical process of decreasing concentrations of nitrate (NO3), called denitrification (Schroth et al., 1998). Decline in concentration of this species seems to be related to a further nitrate reduction process. The large increase in the species indicates the infiltration or recharge of water with high levels of NO3. Factor 5 has a large positive loading for pH and negative loading for ORP. Globally, the increase in pH is related to the iron reduction that is widespread in this study area (Borden et al., 1995).

On the basis of the above analyses, the dominant mechanisms or ongoing processes affecting the distribution of the geochemical parameters in May are various geochemical reactions that degrade petroleum hydrocarbons, such as aerobic respiration, nitrate reduction, manganese reduction, iron reduction, and sulfate reduction. In this study, information on methanogenesis was not available because of a deficiency of data on CH4 in the ground water. However, elevated concentrations of CH4 gas in the soils around hot source areas indicate that methanogenesis is occurring at this site (Lee, 2000; Lee et al., 2000).

The loading patterns of the factors extracted from the data for the September sampling are different from those for May. Factor 1 is characterized by high positive loadings for Ca, Mg, Mn, Si, alkalinity, and EC, and a moderate loading for Fe. Factor 2 is associated with high levels of K, Na, SO4, DO, and ORP. Factor 3 has a high positive loading for pH and Fe, and a moderate negative loading for ORP. Factors 4 and 5 have high positive loadings for NO3 and temperature, respectively. As discussed above, the geochemistry in September seems to be affected by water recharge and further degradation reactions (Cheon, 2000; Lee, 2000; Lee et al., 2001a,b). Therefore, the factors can be interpreted in the context of dispersion and/or dilution, which are important mechanisms for natural attenuation (McAllister and Chiang, 1994), and a biodegradation effect. In the upgradient wells, the increase or decrease in the geochemical parameters may be strongly related to either direct infiltration or recharge from an upper area of water with high levels of those parameters. However, the increase or decrease of those parameters in the downgradient wells are associated with the transport of those species from the most reduced zone (or dilution effects) and/or further biodegradation. This is because large portions of the downgradient area are paved, preventing direct infiltration (see Fig. 5). In this context, high positive loadings of Factor 1 for the upgradient wells (Fig. 8) indicate an external source of solutes. This could be from an upper area, or water infiltrating directly through the upper soil, or, to a certain extent, a further degradation process (such as iron reduction or manganese reduction) of the petroleum hydrocarbons that increases Fe, Mn, and alkalinity. High positive loadings for DO, SO4, and ORP of Factor 2 are associated with the shallow water recharge from rainfall events, which can supply oxygen, nitrate, and sulfate (Cozzarelli et al., 1995). This is true also for Factor 4. Factor 3 seems to be related to a degradation process, which increases Fe and pH, and decreases ORP. The combined changes in these parameters can be explained by transport, dispersion, or dilution by the recharged water, and biodegradation.



View larger version (42K):
[in this window]
[in a new window]
  		Fig. 8. Areal distribution of factor scores for Factor 1 for May.

 
Cluster Analysis
Cluster analysis was performed to split the water sampling points into a finite number of groups (zones) with similar hydrogeochemical composition or redox state (level). Figures 9a and b show the resulting dendrograms for the May and September ground water samples. Three main groups were identified for both sampling events, based on the overall structure of the dendrograms, the regional distribution, and the geochemical interpretability. The discriminant analysis guaranteed the appropriateness of the clustering. No misclassified observations were detected at a significance level of 0.05. Some of the monitoring wells in each sampling period, namely HM1, LM11, and S23 for May, and HM1 and M2 for September, could not be assigned to any of the three groups. This could be the result of very localized geochemical variations, or other external sources of solutes. Well HM1 is upgradient of the main source location and is slightly, or little, affected by the contamination. Therefore, it showed totally different ground water chemistry. It was primarily influenced by rainfall recharge and/or regional recharge from more upgradient areas. LM11 and S23 can be affected, to some extent, by local flow (from SE to NW or M5 to S23), which is in a different flow direction (see Fig. 3a,b). However, for the present, this effect cannot be readily quantified. To analyze this affect, numerical modeling is required (Lee et al., 2001b). Several distinct zones, where different levels of oxidation–reduction processes dominate, were identified (Fig. 10). The hydrogeochemical parameters characterizing each of these zones are summarized in Table 5.



View larger version (17K):
[in this window]
[in a new window]
  		Fig. 9. Dendrograms formed using factor scores for (a) May 1999 and (b) September 1999.

 


View larger version (58K):
[in this window]
[in a new window]
  		Fig. 10. Results of the clustering for (a) May and (b) September 1999.

 

View this table:
[in this window]
[in a new window]
  		Table 5. Statistics of degradation indicators of different groups classified in the cluster analyses.

 
From Table 5, it can be seen that Zones A, B, and C exhibit increasing levels in the overall redox state. The changes in the magnitudes or concentrations of these degradation parameters, relative to reference levels, indicate that a very large amount of biodegradation is occurring at the site. The alkalinity within the downgradient zone of TEX contamination is higher than in the reference ground water. High alkalinity within the TEX plume is expected because biodegradation of the TEX generates aqueous CO2, and thus the alkalinity should increase (Cozzarelli et al., 1995; Basberg et al., 1998). Depressed levels of sulfate and oxidation–reduction potential (ORP) are often observed within and downgradient of the source area, compared with those upgradient of HM1. Elevated concentrations of dissolved iron (Fe) and manganese (Mn) in the downgradient wells, relative to HM1, indicate that iron and manganese reduction occurred in this area (Kennedy et al., 1999).

Pumping Effects
It is very interesting to note that the pattern of zonation parallel to the main direction of ground water flow (to the northwest) was unexpected, because redox zonation is generally developed in a direction perpendicular to ground water flow from the source area (Bjerg et al., 1995; Rügge et al., 1995; Eganhouse et al., 1996; Christensen et al., 2000). The expectation that Zone A, closer to the source area, would be more reduced than Zone B was also rejected by the observed indicator parameters (see Table 5). The most plausible explanation for the abnormal distributions of degradation indicator parameters is the effect of the pump-and-treatment action over the preceding 4 yr. As previously stated, pumping occurred in the W-series wells and was stopped in August 1998, immediately before the initiation of this study. Pumping at W3, W4, W5, and W6 wells continued until August 1998, while pumping at W1 and W2 wells ceased in June 1996. Therefore, we could expect that much of the dissolved hydrocarbons and free products passed the central area of the study site rapidly and then migrated toward the area of the pumping wells (W3–W6). The subsequent evolution of the ground water during pump-and-treatment and after pumping cessation in this area seems to be well explained by the results of the above cluster analyses. In summary, the distribution of the hydrogeochemical parameters in May 1999 was influenced by the pumping at the wells, and subsequent and/or concurrent indigenous biodegradation reactions.

Despite the above analyses, the distribution of the biodegradation indicator parameters at the time of sampling (May, in this case) might not represent the "real" redox conditions at the specific sampling points. This is because certain amounts of these species or parameters had been transported from other areas. The possibility of the facilitated transport of the chemical species is partly supported by the substantial increases in dissolved iron (Fe) at downgradient wells in September for short time intervals (see Fig. 5b). In the interpretation of the clustering results for the May sampling data, we implicitly assumed that the present (at the time of sampling) values of biodegradation-related parameters were mainly derived from indigenous biodegradation processes at the sampling points. However, in considering the above transport process, the redox zonation apparent in Fig. 10 may result from both biodegradation and transport. At the present time, the relative contribution of these effects cannot be readily quantified. For the quantitative evaluation of transport enhancement by pumping, modeling efforts using an appropriate model are required.

Mixing–Dilution and Pavement Effects
In September the distribution of the geochemical parameters obtained from the cluster analysis (Fig. 10b) was different from that in May. The zonation pattern is similar, but Zone C in May was incorporated into Zone B in September. The incorporation of these wells (HM2, 3, and 4) into Zone B indicates that a homogenizing process had occurred during the sampling interval in this area. The process is highly associated with mixing or dilution in upgradient wells, with water carrying high or low levels of the hydrogeochemical parameters from the more upgradient area. The nearly uniform decline in concentrations of total TEX in most wells (by an average of 35%) can be explained by the mixing process (Lee et al., 2000). Zone A in May differs little from Zone A in September, except that well M2 replaced LM11 in the September measurements. Overall inspection of geochemical data in September, relative to those in May, reveals that most of the decrease and increase in parameters conforms to the notion of newly supplied water due to rainfall during the wet season (June to August). The distribution of the geochemical parameters in this period cannot be totally attributed to indigenous biodegradation reactions.

Irrespective of the above interpretation considering the key role of rainfall recharge, it is very interesting to note that the upgradient area was directly affected by the newly recharged water, while the downgradient area might not be influenced substantially by the recharge. As mentioned previously, most of the downgradient area (i.e., textile company) was paved. Therefore, the area did not receive direct rainfall infiltration. The geochemical and/or biological evolution of ground water in the downgradient area in September can be understood by the further biodegradation. The changes in redox indicators along the main flow path can support this interpretation (see Fig. 5). In upgradient wells, concentrations of NO3, SO4, and ORP increased due to the newly recharged water, while those in downgradient wells decreased. The decrease of iron in upgradient wells can be caused by reprecipitation due to the large influx of O2–rich recharge, and the increase of iron in downgradient wells may be due to further iron reduction. The geochemical behavior in the downgradient area can be partly understood by further biodegradation processes such as aerobic respiration, denitrification, and iron and sulfate reduction.

From the above cluster analyses and considerations, it can be inferred that the distribution of the geochemical parameters in May was predominantly affected by the antecedent long-term pumping. The distribution in September was essentially dominated by the mixing–dilution of the recharged water due to rainfall, especially in the upgradient area, and/or further biodegradation in the downgradient area.


   CONCLUSIONS AND DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 CONCLUSIONS AND DISCUSSION
REFERENCES
 
Multivariate analyses (factor and cluster analyses) were applied to the geochemical data of a ground water system contaminated with petroleum hydrocarbons. From the factor analyses, we have identified several geochemical processes that control the composition or redox conditions of ground water. The indigenous biodegradation processes include aerobic respiration, nitrate reduction, iron and manganese reduction, and sulfate reduction. These biodegradation processes were inferred from the highly depressed levels of dissolved oxygen, nitrate, and sulfate, and elevated concentrations of dissolved iron and manganese in ground water, relative to reference levels. However, due to a large amount of rain during the wet season, the geochemistry of the ground water in September was affected by dilution and mixing with newly recharged water, or vertical infiltration. The cluster analyses revealed that the distribution of the geochemical parameters in May was highly associated with the pump-and-treatment remediation action over the preceding 4 yr, and subsequent or concurrent indigenous biodegradation reactions. Piper plots of the major parameters showed a different classification result from the cluster analyses. The Piper plots did not effectively reveal the effects of the pump-and-treatment in the area. This is mainly due to the limited number of parameters used. However, the Piper plot clearly represented the mixing or homogenizing process due to the recharged water.

A kriging method appears very appropriate in this kind of research work. The results are then contours of kriged lines of raw values of water chemistry and/or factor scores (e.g., Subbarao et al., 1996; Carlon et al., 2001) instead of cluster groups. However, sometimes using the kriging technique to present the distribution of each variable or parameter would require a large number of figures, which might not efficiently explain overall water chemistry. More importantly, we are essentially intending to demonstrate how multivariate analysis or the clustering method can be efficiently used in this kind of work.

We have used a limited number of the ground water sampling points. For more reliable and stable clustering (grouping) of ground water, and identification of possible underlying geochemical heterogeneity within the study area, more ground water sampling points are needed. In addition, for an investigation of the vertical variability in the redox conditions within the ground water column at each point, some multilevel sampling is required.


   ACKNOWLEDGMENTS
 
Korea Science and Engineering Foundation partially supported this study to the third author through the project 2000-2-13100-001-1. The Brain Korea 21 program provided a student research fellowship to the first author. We are greatly indebted to Dr. Dirk Mallants and two anonymous reviewers for their valuable comments.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 CONCLUSIONS AND DISCUSSION
 REFERENCES
 




This article has been cited by other articles:


Home page
 J. Environ. Qual.Home page
K. G. Wayland, D. T. Long, D. W. Hyndman, B. C. Pijanowski, S. M. Woodhams, and S. K. Haack
Identifying Relationships between Baseflow Geochemistry and Land Use with Synoptic Sampling and R-Mode Factor Analysis
J. Environ. Qual., January 1, 2003; 32(1): 180 - 190.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via ISI Web of Science (14)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Lee, J.-Y.
Right arrow Articles by Lee, M.-H.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Lee, J.-Y.
Right arrow Articles by Lee, M.-H.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Lee, J.-Y.
Right arrow Articles by Lee, M.-H.
Related Collections
Right arrow Ground Water Quality
Right arrow Water Quality
Right arrow Water Pollution

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Vadose Zone Journal Journal of Plant Registrations
Journal of Natural Resources
and Life Sciences Education		 Soil Science Society of America Journal